SenseNova-Skills: Office Agent Skill Suite
Modular SenseNova skills for building AI-powered office assistants — covering PPT generation, Excel data analysis, infographics, and deep research, usable standalone or composed into end-to-end workflows.
Evidence shows: skills declare triggers and capabilities via SKILL.md, but no least-privilege details; some skills (e.g., sn-ppt-entry) ask user to choose mode, but not all external actions require user confirmation; data flow transparency is limited, no detailed data handling/transmission; sensitive data handling not specified; dependencies not pinned, supply chain risk; external effects (web search, file writes) not explicitly restricted; no rollback mechanism; source attribution partially clear (e.g., citations), but overall source tracking incomplete. Deductions: lack of concrete evidence for least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects control, rollback.
Evidence shows: skill descriptions consistent with README, but no tests or verification; dependencies not pinned, availability uncertain; error handling documentation limited, no detailed failure messages. Deductions: dependency availability unverified, failure messages insufficient.
Evidence shows: README clearly identifies target audience (AI office assistants) and scenarios (image, PPT, data analysis, deep research); capability boundaries described via SKILL.md, but triggers not precise; environment fit (OpenClaw, hermes-agent) documented. Deductions: trigger precision and boundary descriptions not precise enough.
Evidence shows: clear information architecture, well-organized skill directories; detailed install notes (INSTALL.md); stable naming; examples and FAQ provided; known limitations not explicitly listed; MIT license clear; versioning/changelog not provided; maintenance responsibility not clear. Deductions: missing known limitations, versioning/changelog, maintenance responsibility.
Evidence shows: high output usability (generates PPT, reports); clear marginal value (end-to-end workflows); cost-benefit not detailed (e.g., API costs). Deductions: insufficient cost-benefit analysis.
Evidence shows: README claims partially supported by examples, but no traceable test results; cross-source verification limited; facts and inferences not clearly separated. Deductions: lack of verifiable tests and cross-source verification.
- Dependencies are not pinned, posing supply chain risk.
- No rollback mechanism provided, making recovery difficult after failures.
- Sensitive data handling and external effects control are unclear.
What does this agent do, and when should you use it?
SenseNova-Skills is a collection of skills from OpenSenseNova, each declared as a SKILL.md file following the Agent Skills convention, defining triggers, capabilities, and execution flows. They plug into agent runtimes like OpenClaw and hermes-agent, extending models with concrete office capabilities. The repository provides five skill categories: Image & Visualization (sn-image-base, sn-infographic, etc.), Presentations/PPT (sn-ppt-entry, sn-ppt-standard, sn-ppt-creative), Data Analysis (sn-da-excel-workflow, etc.), Deep Research (sn-deep-research, etc.), and Multi-platform Search (academic, code, social). Skills depend on SenseNova Platform API and models (e.g., SenseNova 6.7, U1), with optional web search fallback (e.g., Serper). They can be used individually or chained into full workflows, as demonstrated by the memory-price end-to-end example. The suite is bundled into the Raccoon office product with zero setup, and the repository is MIT-licensed.
Each skill in the repository declares triggers and flow in a SKILL.md file. Concrete operations include: sn-image-base provides low-level tools for text-to-image (sn-image-generate), image recognition (sn-image-recognize), and text optimization (sn-text-optimize) via a unified sn_agent_runner.py; sn-infographic uses the U1 model to generate infographics with prompt-quality scoring, layout/style selection (87 layouts / 66 styles), and multi-round VLM review; sn-ppt-entry parses uploaded PDF/DOCX/MD/TXT, emits task_pack.json and info_pack.json, and dispatches to fast, standard, or creative modes — standard mode builds per-page HTML, runs VLM QC, and exports PPTX; sn-da-excel-workflow reads multi-sheet Excel files, detects large files (≥10k rows triggers Parquet conversion), cleans, filters, cross-sheet aggregates, and exports Excel/CSV; sn-deep-research orchestrates multi-stage research with DAG dependencies, source snapshots, and evidence-informed content units, producing a final report.md (convertible to HTML). Skills also include environment doctors (sn-image-doctor, sn-ppt-doctor) that check dependencies and fill .env, and search aggregators (sn-search-academic, sn-search-code, sn-search-social-cn/en).
- An office worker needs to quickly generate a slide deck with charts and data visuals from a text brief.
- A data analyst must combine multiple Excel files, clean data, and perform cross-sheet aggregation, then export results for reporting.
- A market researcher wants to conduct deep research from web sources and produce a structured report with citations.
- A designer needs to generate a new image imitating a reference or create a resume image from provided info.
- A product manager wants to turn complex data into an infographic for easy sharing.
- A developer wants to integrate SenseNova models into their own agent workflows for office automation and end-to-end pipelines.
What are this agent's strengths and limitations?
- Modular design allows standalone or composed end-to-end workflows covering core office tasks.
- Rich examples (e.g., memory-price end-to-end analysis) showcase full data-to-PPT pipelines.
- Built-in doctor skills automate environment checks, reducing setup friction.
- Infographic generation offers 87 layouts and 66 styles with VLM quality review for professional output.
- Deep research uses DAG-based dependency management for traceable evidence.
- Strong lock-in to SenseNova Platform API and models; porting to other providers requires adaptation.
- Requires configuration of multiple third-party service keys (SenseNova, Serper), making initial setup complex.
- Documentation is fragmented across multiple docs files, with no single consolidated installation guide.
- Chinese social search skills (e.g., Bilibili) require cookie auth, limiting usability.
- No Docker or one-click deployment; users must manage Python and Node.js environments manually.
How do you install or deploy this agent?
Prerequisites: clone the repo and copy skills/ subdirectories into your agent's skills directory. Recommended runtimes: OpenClaw or hermes-agent. For OpenClaw: git clone https://github.com/OpenSenseNova/SenseNova-Skills.git --depth=1; mkdir -p ~/.openclaw/skills; cp -r SenseNova-Skills/skills/* ~/.openclaw/skills/. For hermes-agent, use ~/.hermes/skills/. You need Python and Node.js (for PPT skills), and configure SenseNova Platform API key (free token plan) as well as search API keys (e.g., Serper) for some skills. Detailed dependencies and configs are in the docs/ guides.
How do you use this agent?
After installation, invoke skills via natural language in your agent. For example, tell OpenClaw: "Please install SenseNova-Skills from https://github.com/OpenSenseNova/SenseNova-Skills into your skills directory" (if not already). For a PPT request like "Create a deck on memory price trends", the agent will call sn-ppt-entry and sn-ppt-standard. Ensure environment variables like SENSENOVA_API_KEY are set; doctor skills can auto-check and fill .env. Alternatively, use the Raccoon product for a zero-setup experience.
How does this agent compare with similar options?
Compared to general-purpose models from OpenAI or Anthropic, SenseNova models via this skill suite offer deep office-task customization. However, the repository does not provide direct comparisons to other office automation tools.